Since the blamed commit, closing the first subflow resets the first
subflow socket state to SS_UNCONNECTED.
The current mptcp listen implementation relies only on such
state to prevent touching not-fully-disconnected sockets.
Incoming mptcp fastclose (or paired endpoint removal) unconditionally
closes the first subflow.
All the above allows an incoming fastclose followed by a listen() call
to successfully race with a blocking recvmsg(), potentially causing the
latter to hit a divide by zero bug in cleanup_rbuf/__tcp_select_window().
Address the issue explicitly checking the msk socket state in
mptcp_listen(). An alternative solution would be moving the first
subflow socket state update into mptcp_disconnect(), but in the long
term the first subflow socket should be removed: better avoid relaying
on it for internal consistency check.
Fixes: b29fcfb54c ("mptcp: full disconnect implementation")
Cc: stable@vger.kernel.org
Reported-by: Christoph Paasch <cpaasch@apple.com>
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/414
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
While tacking care of the mptcp-level listener I unintentionally
moved the subflow level unhash after the subflow listener backlog
cleanup.
That could cause some nasty race and makes the code harder to read.
Address the issue restoring the proper order of operations.
Fixes: 57fc0f1cea ("mptcp: ensure listener is unhashed before updating the sk status")
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
The MPTCP code always set the msk state to TCP_CLOSE before
calling performing the fast-close. Move such state transition in
mptcp_do_fastclose() to avoid some code duplication.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The user-space need to properly account the data received/sent by
individual subflows. When additional subflows are created and/or
closed during the MPTCP socket lifetime, the information currently
exposed via MPTCP_TCPINFO are not enough: subflows are identified only
by the sequential position inside the info dumps, and that will change
with the above mentioned events.
To solve the above problem, this patch introduces a new subflow
identifier that is unique inside the given MPTCP socket scope.
The initial subflow get the id 1 and the other subflows get incremental
values at join time.
Link: https://github.com/multipath-tcp/mptcp_net-next/issues/388
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Currently there are no data transfer counters accounting for all
the subflows used by a given MPTCP socket. The user-space can compute
such figures aggregating the subflow info, but that is inaccurate
if any subflow is closed before the MPTCP socket itself.
Add the new counters in the MPTCP socket itself and expose them
via the existing diag and sockopt. While touching mptcp_diag_fill_info(),
acquire the relevant locks before fetching the msk data, to ensure
better data consistency
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/385
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
That will avoid an unneeded conditional in both the fast-path
and in the fallback case and will simplify a bit the next
patch.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The MPTCP protocol access the listener subflow in a lockless
manner in a couple of places (poll, diag). That works only if
the msk itself leaves the listener status only after that the
subflow itself has been closed/disconnected. Otherwise we risk
deadlock in diag, as reported by Christoph.
Address the issue ensuring that the first subflow (the listener
one) is always disconnected before updating the msk socket status.
Reported-by: Christoph Paasch <cpaasch@apple.com>
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/407
Fixes: b29fcfb54c ("mptcp: full disconnect implementation")
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Thanks to the previous patch -- "mptcp: consolidate fallback and non
fallback state machine" -- we can finally drop the "temporary hack"
used to detect rx eof.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
An orphaned msk releases the used resources via the worker,
when the latter first see the msk in CLOSED status.
If the msk status transitions to TCP_CLOSE in the release callback
invoked by the worker's final release_sock(), such instance of the
workqueue will not take any action.
Additionally the MPTCP code prevents scheduling the worker once the
socket reaches the CLOSE status: such msk resources will be leaked.
The only code path that can trigger the above scenario is the
__mptcp_check_send_data_fin() in fallback mode.
Address the issue removing the special handling of fallback socket
in __mptcp_check_send_data_fin(), consolidating the state machine
for fallback and non fallback socket.
Since non-fallback sockets do not send and do not receive data_fin,
the mptcp code can update the msk internal status to match the next
step in the SM every time data fin (ack) should be generated or
received.
As a consequence we can remove a bunch of checks for fallback from
the fastpath.
Fixes: 6e628cd3a8 ("mptcp: use mptcp release_cb for delayed tasks")
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
At passive MPJ time, if the msk socket lock is held by the user,
the new subflow is appended to the msk->join_list under the msk
data lock.
In mptcp_release_cb()/__mptcp_flush_join_list(), the subflows in
that list are moved from the join_list into the conn_list under the
msk socket lock.
Append and removal could race, possibly corrupting such list.
Address the issue splicing the join list into a temporary one while
still under the msk data lock.
Found by code inspection, the race itself should be almost impossible
to trigger in practice.
Fixes: 3e5014909b ("mptcp: cleanup MPJ subflow list handling")
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Currently the mptcp code has assumes that disconnect() can fail only
at mptcp_sendmsg_fastopen() time - to avoid a deadlock scenario - and
don't even bother returning an error code.
Soon mptcp_disconnect() will handle more error conditions: let's track
them explicitly.
As a bonus, explicitly annotate TCP-level disconnect as not failing:
the mptcp code never blocks for event on the subflows.
Fixes: 7d803344fd ("mptcp: fix deadlock in fastopen error path")
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Tested-by: Christoph Paasch <cpaasch@apple.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Most of the ioctls to net protocols operates directly on userspace
argument (arg). Usually doing get_user()/put_user() directly in the
ioctl callback. This is not flexible, because it is hard to reuse these
functions without passing userspace buffers.
Change the "struct proto" ioctls to avoid touching userspace memory and
operate on kernel buffers, i.e., all protocol's ioctl callbacks is
adapted to operate on a kernel memory other than on userspace (so, no
more {put,get}_user() and friends being called in the ioctl callback).
This changes the "struct proto" ioctl format in the following way:
int (*ioctl)(struct sock *sk, int cmd,
- unsigned long arg);
+ int *karg);
(Important to say that this patch does not touch the "struct proto_ops"
protocols)
So, the "karg" argument, which is passed to the ioctl callback, is a
pointer allocated to kernel space memory (inside a function wrapper).
This buffer (karg) may contain input argument (copied from userspace in
a prep function) and it might return a value/buffer, which is copied
back to userspace if necessary. There is not one-size-fits-all format
(that is I am using 'may' above), but basically, there are three type of
ioctls:
1) Do not read from userspace, returns a result to userspace
2) Read an input parameter from userspace, and does not return anything
to userspace
3) Read an input from userspace, and return a buffer to userspace.
The default case (1) (where no input parameter is given, and an "int" is
returned to userspace) encompasses more than 90% of the cases, but there
are two other exceptions. Here is a list of exceptions:
* Protocol RAW:
* cmd = SIOCGETVIFCNT:
* input and output = struct sioc_vif_req
* cmd = SIOCGETSGCNT
* input and output = struct sioc_sg_req
* Explanation: for the SIOCGETVIFCNT case, userspace passes the input
argument, which is struct sioc_vif_req. Then the callback populates
the struct, which is copied back to userspace.
* Protocol RAW6:
* cmd = SIOCGETMIFCNT_IN6
* input and output = struct sioc_mif_req6
* cmd = SIOCGETSGCNT_IN6
* input and output = struct sioc_sg_req6
* Protocol PHONET:
* cmd == SIOCPNADDRESOURCE | SIOCPNDELRESOURCE
* input int (4 bytes)
* Nothing is copied back to userspace.
For the exception cases, functions sock_sk_ioctl_inout() will
copy the userspace input, and copy it back to kernel space.
The wrapper that prepare the buffer and put the buffer back to user is
sk_ioctl(), so, instead of calling sk->sk_prot->ioctl(), the callee now
calls sk_ioctl(), which will handle all cases.
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20230609152800.830401-1-leitao@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Active subflow are inserted into the connection list at creation time.
When the MPJ handshake completes successfully, a new subflow creation
netlink event is generated correctly, but the current code wrongly
avoid initializing a couple of subflow data.
The above will cause misbehavior on a few exceptional events: unneeded
mptcp-level retransmission on msk-level sequence wrap-around and infinite
mapping fallback even when a MPJ socket is present.
Address the issue factoring out the needed initialization in a new helper
and invoking the latter from __mptcp_finish_join() time for passive
subflow and from mptcp_finish_join() for active ones.
Fixes: 0530020a7c ("mptcp: track and update contiguous data status")
Cc: stable@vger.kernel.org
Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Christoph reported the mptcp variant of a recently addressed plain
TCP issue. Similar to commit e14cadfd80 ("tcp: add annotations around
sk->sk_shutdown accesses") add READ/WRITE ONCE annotations to silence
KCSAN reports around lockless sk_shutdown access.
Fixes: 71ba088ce0 ("mptcp: cleanup accept and poll")
Reported-by: Christoph Paasch <cpaasch@apple.com>
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/401
Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The first subflow socket is accessed outside the msk socket lock
by mptcp_subflow_fail(), we need to annotate each write access
with WRITE_ONCE, but a few spots still lacks it.
Fixes: 76a13b3157 ("mptcp: invoke MP_FAIL response when needed")
Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When the msk socket is cloned at MPC handshake time, a few
fields are initialized in a racy way outside mptcp_sk_clone()
and the msk socket lock.
The above is due historical reasons: before commit a88d0092b2
("mptcp: simplify subflow_syn_recv_sock()") as the first subflow socket
carrying all the needed date was not available yet at msk creation
time
We can now refactor the code moving the missing initialization bit
under the socket lock, removing the init race and avoiding some
code duplication.
This will also simplify the next patch, as all msk->first write
access are now under the msk socket lock.
Fixes: 0397c6d85f ("mptcp: keep unaccepted MPC subflow into join list")
Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The MPTCP can access the first subflow socket in a few spots
outside the socket lock scope. That is actually safe, as MPTCP
will delete the socket itself only after the msk sock close().
Still the such accesses causes a few KCSAN splats, as reported
by Christoph. Silence the harmless warning adding a few annotation
around the relevant accesses.
Fixes: 71ba088ce0 ("mptcp: cleanup accept and poll")
Reported-by: Christoph Paasch <cpaasch@apple.com>
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/402
Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Ondrej reported a functional issue WRT timeout handling on connect
with a nice reproducer.
The problem is that the current mptcp connect waits for both the
MPTCP socket level timeout, and the first subflow socket timeout.
The latter is not influenced/touched by the exposed setsockopt().
Overall the above makes the SO_SNDTIMEO a no-op on connect.
Since mptcp_connect is invoked via inet_stream_connect and the
latter properly handle the MPTCP level timeout, we can address the
issue making the nested subflow level connect always unblocking.
This also allow simplifying a bit the code, dropping an ugly hack
to handle the fastopen and custom proto_ops connect.
The issues predates the blamed commit below, but the current resolution
requires the infrastructure introduced there.
Fixes: 54f1944ed6 ("mptcp: factor out mptcp_connect()")
Reported-by: Ondrej Mosnacek <omosnace@redhat.com>
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/399
Cc: stable@vger.kernel.org
Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Rewrite the mptcp socket accept op, leveraging the new
__inet_accept() helper.
This way we can avoid acquiring the new socket lock twice
and we can avoid a couple of indirect calls.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When cleaning up unaccepted mptcp socket still laying inside
the listener queue at listener close time, such sockets will
go through a regular close, waiting for a timeout before
shutting down the subflows.
There is no need to keep the kernel resources in use for
such a possibly long time: short-circuit to fast-close.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the long run this will simplify the mptcp code and will
allow for more consistent behavior. Move the first subflow
allocation out of the sock->init ops into the __mptcp_nmpc_socket()
helper.
Since the first subflow creation can now happen after the first
setsockopt() we additionally need to invoke mptcp_sockopt_sync()
on it.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
So that we can avoid a bunch of check in fastpath. Additionally we
can specialize such check according to the specific fastopen method
- defer_connect vs MSG_FASTOPEN.
The latter bits will simplify the next patches.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
In a few spots, the mptcp code invokes the __mptcp_nmpc_socket() helper
multiple times under the same socket lock scope. Additionally, in such
places, the socket status ensures that there is no MP capable handshake
running.
Under the above condition we can replace the later __mptcp_nmpc_socket()
helper invocation with direct access to the msk->subflow pointer and
better document such access is not supposed to fail with WARN().
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
We can change mptcp_sk() to propagate its argument const qualifier,
thanks to container_of_const().
We need to change few things to avoid build errors:
mptcp_set_datafin_timeout() and mptcp_rtx_head() have to accept
non-const sk pointers.
@msk local variable in mptcp_pending_tail() must be const.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Matthieu Baerts <matthieu.baerts@tessares.net>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
mptcp_poll() reads sk->sk_err without socket lock held/owned.
Add READ_ONCE() and WRITE_ONCE() to avoid load/store tearing.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As reported by Christoph after having refactored the passive
socket initialization, the mptcp listener shutdown path is prone
to an UaF issue.
BUG: KASAN: use-after-free in _raw_spin_lock_bh+0x73/0xe0
Write of size 4 at addr ffff88810cb23098 by task syz-executor731/1266
CPU: 1 PID: 1266 Comm: syz-executor731 Not tainted 6.2.0-rc59af4eaa31c1f6c00c8f1e448ed99a45c66340dd5 #6
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
Call Trace:
<TASK>
dump_stack_lvl+0x6e/0x91
print_report+0x16a/0x46f
kasan_report+0xad/0x130
kasan_check_range+0x14a/0x1a0
_raw_spin_lock_bh+0x73/0xe0
subflow_error_report+0x6d/0x110
sk_error_report+0x3b/0x190
tcp_disconnect+0x138c/0x1aa0
inet_child_forget+0x6f/0x2e0
inet_csk_listen_stop+0x209/0x1060
__mptcp_close_ssk+0x52d/0x610
mptcp_destroy_common+0x165/0x640
mptcp_destroy+0x13/0x80
__mptcp_destroy_sock+0xe7/0x270
__mptcp_close+0x70e/0x9b0
mptcp_close+0x2b/0x150
inet_release+0xe9/0x1f0
__sock_release+0xd2/0x280
sock_close+0x15/0x20
__fput+0x252/0xa20
task_work_run+0x169/0x250
exit_to_user_mode_prepare+0x113/0x120
syscall_exit_to_user_mode+0x1d/0x40
do_syscall_64+0x48/0x90
entry_SYSCALL_64_after_hwframe+0x72/0xdc
The msk grace period can legitly expire in between the last
reference count dropped in mptcp_subflow_queue_clean() and
the later eventual access in inet_csk_listen_stop()
After the previous patch we don't need anymore special-casing
msk listener socket cleanup: the mptcp worker will process each
of the unaccepted msk sockets.
Just drop the now unnecessary code.
Please note this commit depends on the two parent ones:
mptcp: refactor passive socket initialization
mptcp: use the workqueue to destroy unaccepted sockets
Fixes: 6aeed90450 ("mptcp: fix race on unaccepted mptcp sockets")
Cc: stable@vger.kernel.org
Reported-and-tested-by: Christoph Paasch <cpaasch@apple.com>
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/346
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Christoph reported a UaF at token lookup time after having
refactored the passive socket initialization part:
BUG: KASAN: use-after-free in __token_bucket_busy+0x253/0x260
Read of size 4 at addr ffff88810698d5b0 by task syz-executor653/3198
CPU: 1 PID: 3198 Comm: syz-executor653 Not tainted 6.2.0-rc59af4eaa31c1f6c00c8f1e448ed99a45c66340dd5 #6
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
Call Trace:
<TASK>
dump_stack_lvl+0x6e/0x91
print_report+0x16a/0x46f
kasan_report+0xad/0x130
__token_bucket_busy+0x253/0x260
mptcp_token_new_connect+0x13d/0x490
mptcp_connect+0x4ed/0x860
__inet_stream_connect+0x80e/0xd90
tcp_sendmsg_fastopen+0x3ce/0x710
mptcp_sendmsg+0xff1/0x1a20
inet_sendmsg+0x11d/0x140
__sys_sendto+0x405/0x490
__x64_sys_sendto+0xdc/0x1b0
do_syscall_64+0x3b/0x90
entry_SYSCALL_64_after_hwframe+0x72/0xdc
We need to properly clean-up all the paired MPTCP-level
resources and be sure to release the msk last, even when
the unaccepted subflow is destroyed by the TCP internals
via inet_child_forget().
We can re-use the existing MPTCP_WORK_CLOSE_SUBFLOW infra,
explicitly checking that for the critical scenario: the
closed subflow is the MPC one, the msk is not accepted and
eventually going through full cleanup.
With such change, __mptcp_destroy_sock() is always called
on msk sockets, even on accepted ones. We don't need anymore
to transiently drop one sk reference at msk clone time.
Please note this commit depends on the parent one:
mptcp: refactor passive socket initialization
Fixes: 58b0991962 ("mptcp: create msk early")
Cc: stable@vger.kernel.org
Reported-and-tested-by: Christoph Paasch <cpaasch@apple.com>
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/347
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
After commit 30e51b923e ("mptcp: fix unreleased socket in accept queue")
unaccepted msk sockets go throu complete shutdown, we don't need anymore
to delay inserting the first subflow into the subflow lists.
The reference counting deserve some extra care, as __mptcp_close() is
unaware of the request socket linkage to the first subflow.
Please note that this is more a refactoring than a fix but because this
modification is needed to include other corrections, see the following
commits. Then a Fixes tag has been added here to help the stable team.
Fixes: 30e51b923e ("mptcp: fix unreleased socket in accept queue")
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Tested-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit e48c414ee6 ("[INET]: Generalise the TCP sock ID lookup routines")
commented out the definition of SOCK_REFCNT_DEBUG in 2005 and later another
commit 463c84b97f ("[NET]: Introduce inet_connection_sock") removed it.
Since we could track all of them through bpf and kprobe related tools
and the feature could print loads of information which might not be
that helpful even under a little bit pressure, the whole feature which
has been inactive for many years is no longer supported.
Link: https://lore.kernel.org/lkml/20230211065153.54116-1-kerneljasonxing@gmail.com/
Suggested-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Jason Xing <kernelxing@tencent.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Acked-by: Wenjia Zhang <wenjia@linux.ibm.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the peer closes all the existing subflows for a given
mptcp socket and later the application closes it, the current
implementation let it survive until the timewait timeout expires.
While the above is allowed by the protocol specification it
consumes resources for almost no reason and additionally
causes sporadic self-tests failures.
Let's move the mptcp socket to the TCP_CLOSE state when there are
no alive subflows at close time, so that the allocated resources
will be freed immediately.
Fixes: e16163b6e2 ("mptcp: refactor shutdown and close")
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Let the caller specify the to-be-created subflow family.
For a given MPTCP socket created with the AF_INET6 family, the current
userspace PM can already ask the kernel to create subflows in v4 and v6.
If "plain" IPv4 addresses are passed to the kernel, they are
automatically mapped in v6 addresses "by accident". This can be
problematic because the userspace will need to pass different addresses,
now the v4-mapped-v6 addresses to destroy this new subflow.
On the other hand, if the MPTCP socket has been created with the AF_INET
family, the command to create a subflow in v6 will be accepted but the
result will not be the one as expected as new subflow will be created in
IPv4 using part of the v6 addresses passed to the kernel: not creating
the expected subflow then.
No functional change intended for the in-kernel PM where an explicit
enforcement is currently in place. This arbitrary enforcement will be
leveraged by other patches in a future version.
Fixes: 702c2f646d ("mptcp: netlink: allow userspace-driven subflow establishment")
Cc: stable@vger.kernel.org
Co-developed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Do the statistics of mptcp socket in use with sock_prot_inuse_add().
Therefore, we can get the count of used mptcp socket from
/proc/net/protocols:
& cat /proc/net/protocols
protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em
MPTCPv6 2048 0 0 no 0 yes kernel y n y y y y y y y y y y n n n y y y n
MPTCP 1896 1 0 no 0 yes kernel y n y y y y y y y y y y n n n y y y n
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
'sock->sk' is used frequently in mptcp_listen(). Therefore, we can
introduce the 'sk' and replace 'sock->sk' with it.
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use the local variable 'net' instead of sock_net() in the functions where
the variable 'struct net *net' has been defined.
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Geliang Tang <geliang.tang@suse.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The helper msk_owned_by_me() is defined in protocol.h, so use it instead
of sock_owned_by_me().
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Geliang Tang <geliang.tang@suse.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
MattB reported a lockdep splat in the mptcp listener code cleanup:
WARNING: possible circular locking dependency detected
packetdrill/14278 is trying to acquire lock:
ffff888017d868f0 ((work_completion)(&msk->work)){+.+.}-{0:0}, at: __flush_work (kernel/workqueue.c:3069)
but task is already holding lock:
ffff888017d84130 (sk_lock-AF_INET){+.+.}-{0:0}, at: mptcp_close (net/mptcp/protocol.c:2973)
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #1 (sk_lock-AF_INET){+.+.}-{0:0}:
__lock_acquire (kernel/locking/lockdep.c:5055)
lock_acquire (kernel/locking/lockdep.c:466)
lock_sock_nested (net/core/sock.c:3463)
mptcp_worker (net/mptcp/protocol.c:2614)
process_one_work (kernel/workqueue.c:2294)
worker_thread (include/linux/list.h:292)
kthread (kernel/kthread.c:376)
ret_from_fork (arch/x86/entry/entry_64.S:312)
-> #0 ((work_completion)(&msk->work)){+.+.}-{0:0}:
check_prev_add (kernel/locking/lockdep.c:3098)
validate_chain (kernel/locking/lockdep.c:3217)
__lock_acquire (kernel/locking/lockdep.c:5055)
lock_acquire (kernel/locking/lockdep.c:466)
__flush_work (kernel/workqueue.c:3070)
__cancel_work_timer (kernel/workqueue.c:3160)
mptcp_cancel_work (net/mptcp/protocol.c:2758)
mptcp_subflow_queue_clean (net/mptcp/subflow.c:1817)
__mptcp_close_ssk (net/mptcp/protocol.c:2363)
mptcp_destroy_common (net/mptcp/protocol.c:3170)
mptcp_destroy (include/net/sock.h:1495)
__mptcp_destroy_sock (net/mptcp/protocol.c:2886)
__mptcp_close (net/mptcp/protocol.c:2959)
mptcp_close (net/mptcp/protocol.c:2974)
inet_release (net/ipv4/af_inet.c:432)
__sock_release (net/socket.c:651)
sock_close (net/socket.c:1367)
__fput (fs/file_table.c:320)
task_work_run (kernel/task_work.c:181 (discriminator 1))
exit_to_user_mode_prepare (include/linux/resume_user_mode.h:49)
syscall_exit_to_user_mode (kernel/entry/common.c:130)
do_syscall_64 (arch/x86/entry/common.c:87)
entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)
other info that might help us debug this:
Possible unsafe locking scenario:
CPU0 CPU1
---- ----
lock(sk_lock-AF_INET);
lock((work_completion)(&msk->work));
lock(sk_lock-AF_INET);
lock((work_completion)(&msk->work));
*** DEADLOCK ***
The report is actually a false positive, since the only existing lock
nesting is the msk socket lock acquired by the mptcp work.
cancel_work_sync() is invoked without the relevant socket lock being
held, but under a different (the msk listener) socket lock.
We could silence the splat adding a per workqueue dynamic lockdep key,
but that looks overkill. Instead just tell lockdep the msk socket lock
is not held around cancel_work_sync().
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/322
Fixes: 30e51b923e ("mptcp: fix unreleased socket in accept queue")
Reported-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
MatM reported a deadlock at fastopening time:
INFO: task syz-executor.0:11454 blocked for more than 143 seconds.
Tainted: G S 6.1.0-rc5-03226-gdb0157db5153 #1
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
task:syz-executor.0 state:D stack:25104 pid:11454 ppid:424 flags:0x00004006
Call Trace:
<TASK>
context_switch kernel/sched/core.c:5191 [inline]
__schedule+0x5c2/0x1550 kernel/sched/core.c:6503
schedule+0xe8/0x1c0 kernel/sched/core.c:6579
__lock_sock+0x142/0x260 net/core/sock.c:2896
lock_sock_nested+0xdb/0x100 net/core/sock.c:3466
__mptcp_close_ssk+0x1a3/0x790 net/mptcp/protocol.c:2328
mptcp_destroy_common+0x16a/0x650 net/mptcp/protocol.c:3171
mptcp_disconnect+0xb8/0x450 net/mptcp/protocol.c:3019
__inet_stream_connect+0x897/0xa40 net/ipv4/af_inet.c:720
tcp_sendmsg_fastopen+0x3dd/0x740 net/ipv4/tcp.c:1200
mptcp_sendmsg_fastopen net/mptcp/protocol.c:1682 [inline]
mptcp_sendmsg+0x128a/0x1a50 net/mptcp/protocol.c:1721
inet6_sendmsg+0x11f/0x150 net/ipv6/af_inet6.c:663
sock_sendmsg_nosec net/socket.c:714 [inline]
sock_sendmsg+0xf7/0x190 net/socket.c:734
____sys_sendmsg+0x336/0x970 net/socket.c:2476
___sys_sendmsg+0x122/0x1c0 net/socket.c:2530
__sys_sendmmsg+0x18d/0x460 net/socket.c:2616
__do_sys_sendmmsg net/socket.c:2645 [inline]
__se_sys_sendmmsg net/socket.c:2642 [inline]
__x64_sys_sendmmsg+0x9d/0x110 net/socket.c:2642
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x38/0x90 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x63/0xcd
RIP: 0033:0x7f5920a75e7d
RSP: 002b:00007f59201e8028 EFLAGS: 00000246 ORIG_RAX: 0000000000000133
RAX: ffffffffffffffda RBX: 00007f5920bb4f80 RCX: 00007f5920a75e7d
RDX: 0000000000000001 RSI: 0000000020002940 RDI: 0000000000000005
RBP: 00007f5920ae7593 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000020004050 R11: 0000000000000246 R12: 0000000000000000
R13: 000000000000000b R14: 00007f5920bb4f80 R15: 00007f59201c8000
</TASK>
In the error path, tcp_sendmsg_fastopen() ends-up calling
mptcp_disconnect(), and the latter tries to close each
subflow, acquiring the socket lock on each of them.
At fastopen time, we have a single subflow, and such subflow
socket lock is already held by the called, causing the deadlock.
We already track the 'fastopen in progress' status inside the msk
socket. Use it to address the issue, making mptcp_disconnect() a
no op when invoked from the fastopen (error) path and doing the
relevant cleanup after releasing the subflow socket lock.
While at the above, rename the fastopen status bit to something
more meaningful.
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/321
Fixes: fa9e57468a ("mptcp: fix abba deadlock on fastopen")
Reported-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch adds two new MPTCP netlink event types for PM listening
socket create and close, named MPTCP_EVENT_LISTENER_CREATED and
MPTCP_EVENT_LISTENER_CLOSED.
Add a new function mptcp_event_pm_listener() to push the new events
with family, port and addr to userspace.
Invoke mptcp_event_pm_listener() with MPTCP_EVENT_LISTENER_CREATED in
mptcp_listen() and mptcp_pm_nl_create_listen_socket(), invoke it with
MPTCP_EVENT_LISTENER_CLOSED in __mptcp_close_ssk().
Signed-off-by: Geliang Tang <geliang.tang@suse.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Paolo Abeni